Загрузка видео...
Не удалось загрузить видео
DeepSeek V4.1 Flash just dropped, and holy shit this thing is fast,I’m getting almost 400 t/s in my video tests. The speed is honestly crazy, I’m running my own benchmark right now and I’ll share all the results once I’m done. You just need to go to the DeepSeek... show more
101,405 просмотров • 3 дней назад •via X (Twitter)
Комментарии: 38
first results, not bad
Yup, just about time, i need deepseek flash run faster
It's been running at that speed for 1 hour, 0.3 cents of dollars!
Why the timer faster 😭
Wait until Watermelon 😏 🍉
I can't wait in a few hours I'll publish a benchmark for Spark 1.3 MAX , I'm letting it run all night
😃 plz tag me, I wanna make sure I see it
400 t/s is wild. The sustained number will be more interesting than the burst.
Thats crazy fast 400 t/s 🫠
why did you speed up the video by 10x?
That’s wild. How’s the memory footprint looking at that speed?
deepseek moving fast and cheap is why spark's 20% fewer tool calls matters. flash isn't trying to beat frontier on hardest tasks, it's trying to be good enough that you stop reaching for the expensive one
If you want to show that it's fast, then why speed up the video? Wouldn't it be better to show what the actual speed feels like?
Speeds like this could change how expensive iteration feels inside agent loops
You got to love how people keep posting things like this and just completely leave out what hardware they're using lol.
Does it have any strengths in math or coding? 🧐
crazy fast
I honestly thought it was running on Cerebras chips. It’s insanely fast, I can’t believe these are Chinese chips.
Crazy …Job I will try tonight
Yes, it's very fast, averaging over 300.
price the same?
Yes !
Holly shit
400 t/s is great until you realize your video is 30 seconds and you still have to wait for the render queue. More importantly is speed that important? Or is accuracy?
Where is it accessible? crazy speed
Neden video hızlandırılmış ki, saçmalık
How’d you get access brother
Just ask Codex, or really any LLM, to configure deepseek-v4.1-flash-expires-on-0910 in your DeepSeek Desktop, That’s it!
Thing is so fast I'll speed it up even more in the videos I post on X so no one can tell how fast it really is in practice
This speed is really too fast. I hope the price can come down some more haha.
400 t/s needs hardware and batch size disclosed, otherwise the deepseek benchmark says little
400 t/s changes the equation for agentic workloads: latency becomes less of a bottleneck than tool-call overhead. The real test is sustained throughput under long context, not peak decode speed.
@SPAC89 worth adding: no v4.1 weights on HF yet — deepseek-ai org tops out at 0731 (4.4M dl). Speed vs Muse Spark 1.3 at /usr/bin/bash.089//usr/bin/bash.177/M is the story; if v4.1 ships a GGUF ladder day-0 like 0731 did, that's the adoption curve to watch.
does not seem available from API
400 t/s is wild. How does the reasoning quality hold up at that speed compared to the standard model?
400 t/s changes how a model feels more than people give it credit for. if quality holds up, a lot of agent workflows benefit more from this kind of latency drop than another small benchmark gain
Dont speed up a speed showcase retarded asshole
Is it multimodal?
![how is this even allowed i dropped my $200/mo claude plan for deepseek harness, it’s basically free the whole thing runs on one simple rule: everything is a plugin [setup takes 3 minutes, here’s the playbook] 1. clone/install the repo called deepseek-harness 2. skim the architecture docs 3. that’s it, you’re live bookmark this and hand it to your agent right now.](https://image.24vids.com/tw-2088952662603731225/amplify_video_thumb/2088781800273231873/img/BKAMULsHEuJBwRpY.jpg)
![DeepSeek V4.1 Flash vs Gemini 3.8 Flash at 3D geometry test: six self-contained three.js stereometry scenes — a (3,8) torus knot, a Klein bottle, a great stellated dodecahedron, a hyperboloid ruled by 120 straight rods, a level-3 Menger sponge, a thick Möbius strip. one shot each, no libraries beyond three.js, every surface painted with a live iridescent rainbow that has to crawl across the geometry. cost: Gemini 3.8 Flash — $0.11 DeepSeek V4.1 Flash — $0.27 • gemini delivers all six figures for 2.4× less — 21k output tokens against deepseek's 168k • gemini is 5× faster — every scene under 25 seconds, deepseek took up to 3 minutes on the Möbius strip • deepseek wins the torus knot: its rope is vivid and glossy, while gemini rendered the same knot nearly black • neither model can build a real Klein bottle. both hand back a twisted tube and call it done try both on aimlapi[.]com](https://image.24vids.com/tw-2098056487465578894/amplify_video_thumb/2098055082444988421/img/BYUIgYB7mmUyE26F.jpg)
